pyro-ppl/pyro

★ 9,001PythonAudience · researcherComplexity · 4/5Setup · moderate

Mindmap

mindmap
  root((repo))
    What it does
      Probabilistic models
      Statistical inference
      Uncertainty modeling
    Tech stack
      Python
      PyTorch
    Use cases
      Bayesian inference
      Deep learning models
      Research models
    Audience
      Researchers
      ML engineers

mindmap root((repo)) What it does Probabilistic models Statistical inference Uncertainty modeling Tech stack Python PyTorch Use cases Bayesian inference Deep learning models Research models Audience Researchers ML engineers

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Build a probabilistic model for medical diagnosis that estimates the probability of a condition given observed test results.

USE CASE 2

Combine a neural network with a statistical prior in a single Pyro model to build a Bayesian deep learning system.

USE CASE 3

Run automatic variational inference on a custom statistical model to fit it to observed data using Pyro's SVI interface.

Tech stack

PythonPyTorchpip

Getting it running

Difficulty · moderate Time to first run · 30min

Requires PyTorch installed first, GPU recommended for larger models but not required.

In plain English

Pyro is a Python library for a style of programming called probabilistic programming. In regular programming, code produces a single definite output for a given input. In probabilistic programming, you write code that describes probability distributions: models of how uncertain quantities relate to each other. The system then figures out, given some observed data, which settings of the unknown quantities best explain what was seen. This matters for problems where you need to reason under uncertainty. A medical diagnostic model might ask: given these test results, what is the probability the patient has condition X? A financial model might ask: given past price movements, what distribution of future returns should we expect? Pyro gives researchers and engineers a way to write those kinds of models in Python and then run automatic statistical inference on them. Pyro is built on PyTorch, a widely used library for deep learning (the technology behind many modern AI systems). This means Pyro models can combine traditional statistical components with neural networks in the same model. The library is designed to scale to large datasets and to be composable: complex models are built by combining simpler pieces. The project was originally developed at Uber's AI research team and is now maintained by community contributors including a team at the Broad Institute, a biomedical research organization. In 2019, it joined the Linux Foundation as an official open source project. It is installed as a Python package via pip and comes with a collection of example models and tutorials covering common use cases. The README includes a citation for the original research paper published in the Journal of Machine Learning Research for those who want to reference the work formally.

Copy-paste prompts

Prompt 1

Help me write a Pyro model that infers the bias of a coin from observed flips using variational inference and ELBO loss.

Prompt 2

Show me how to define a Pyro model with a Gaussian mixture and run MCMC to estimate the posterior distribution.

Prompt 3

I want to build a time-series forecasting model in Pyro that accounts for uncertainty in the predictions. Walk me through the structure.

Prompt 4

Help me combine a PyTorch neural network encoder with a Pyro latent variable model to build a variational autoencoder.

Open on GitHub → Explain another repo

← pyro-ppl on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.